The existing behavior of mongos is to not propagate errors with the RetryableWriteError label, and instead retry internally. In this case, because the drivers cannot reliably maintain error handling logic for errors labeled with NoWritesPerfomed (that is drivers have no access to the original error), the internal behavior of mongos should be modified to meet the following condition:

If a retry attempt fails with an error labeled NoWritesPerformed, mongos MUST return the original error.